Experimental writing at the edge of security research. Tracing assumptions, questioning what isn't mainstream.
Exploit primitives, memory corruption, and kernel internals.
Mechanistic theory and circuit analysis of LLMs.
Formal methods and safety invariants in distributed systems.
A detailed investigation into a race condition discovered in the Linux kernel graphics driver disposal logic, examining the transition from reference counting to memory reclamation.
A mechanistic theory of multi-head attention in transformer architectures, proposing that individual heads act as discrete gates in a larger computational circuit.
Observations on using formal specifications to verify leader election safety and log matching properties in Raft clusters.